A recognition method of mushroom mycelium varieties based on near-infrared spectroscopy and deep learning model

Abstract

Rapid detection of the mycelium varieties of the edible fungi is of great significance for its quality breeding and cultivation. Artificial observation method is inaccurate to judge the mycelium varieties, thus, it is urgent to research efficient detection technology for mycelium varieties. In this study, a recognition method of mycelium varieties was proposed based on near-infrared spectroscopy and deep learning model. First, the near infrared spectral data of six varieties of mushroom mycelium were acquired by Fourier transform near infrared spectrometer. Second, the wavelet packet threshold denoising method and the standard normal variable transformation (SNV) method were used to preprocess the near-infrared (NIR) spectral data of mushroom mycelium. The characteristic wavelength of the preprocessed spectral data was extracted by successive projections algorithm (SPA). In addition, 864 groups of samples after preprocessing were randomly divided into 691 training sets and 173 test sets according to the ratio of 4:1. Finally, based on the 16 wavenumber variables extracted by the SPA, a recognition model of mushroom mycelium varieties was constructed using an eight-layer convolutional neural network (E-CNN). The result showed that an accurate and rapid method for the recognition of edible fungi’s mycelium varieties was realized, with the recognition accuracy of 98.27 % and the running time of 0.000329 s.